ultrareview crashed before producing findings — burned a free run (PR mode, 282 files)

Summary

/ultrareview in PR mode crashed during the Find stage (~10s after start, only 2 heartbeats), returning {"error":"Review crashed before producing findings."}. Free credit was consumed (2/3 → 3/3) despite no usable output.

Environment

• Claude Code CLI: Opus 4.7 (1M context)
• Platform: macOS (Darwin 25.4.0)
• Plan: Pro
• ultrareview mode: PR (/ultrareview 4)

Reproduction

1. Opened PR for a full-codebase audit on a personal project:

https://github.com/dhscfh/stgfolio/pull/4

• 282 changed files / +31,744 / -1,355 lines
• Mix of TS / TSX / Rust / Markdown / JSON / config

1. Ran /ultrareview 4 from CLI
2. Confirmation: Free ultrareview 2 of 3. Ultrareview launched for dhscfh/stgfolio#4
3. After ~5 min, task-notification arrived with status completed
4. Returned payload (verbatim):

``
{"error":"Review crashed before producing findings. See session logs for details."}
``

1. Free quota counter: 2/3 → 3/3

Session log

Session URL: https://claude.ai/code/session_01F9qt1ccEPw85fR2zvZ35bx

Stages observed:

• Setup ✓
• Find ✗ (only 2 heartbeats: [heartbeat] 05:00:55, [heartbeat] 05:01:05 then crash)
• Verify / Dedupe never started

Crashing this fast (~10s) suggests a parser / PR fetch failure, not time/memory exhaustion.

Why I'm filing despite likely duplicate of #55031

Adding for visibility on:

• Smaller scope (282 files vs e.g. #50029's 3318 files which at least ran and returned empty)
• Specific failure mode: explicit crash in Find stage, not silent empty result
• Confirms widespread across PR sizes / repo types

Related: #55031 · #54871 · #54812 · #54516 · #53455 · #53301 · #52780 · #50029 · #49882

Expected

One of:

1. Successful review with findings, OR
2. Clear error: scope exceeds ultrareview limits (max N files / M lines) if size cap hit, OR
3. Free quota NOT consumed when server-side crashes occur (per existing reports, Anthropic has refunded these in similar cases)

Request

Refund the free run consumed by session_01F9qt1ccEPw85fR2zvZ35bx. Happy to provide additional logs if it helps narrow down the parser failure.

Thanks!